Portable, Extensible Toolkit for Scientific Computation (PETSc, https://petsc.org) is a suite of data structures and routines for the scalable (parallel) solution of scientific applications. Due to its solid mathematical grounding, careful software design, and most importantly, evolution resulting from the usage of many users in various application areas, PETSc has enabled engineers and scientists to solve large-scale problems, with previously unreachable resolution, in areas as diverse as river, power and traffic networks, cardiology, fusion, nuclear energy, astrophysics, and climate change.

 

As a PETSc developer, I will first give a brief introduction and tutorial of PETSc library, illustrate how it can be used to facilitate academic research in general. Then I will present a relatively new class in the PETSc library, namely DMNetwork. Work in these thrust centers on creating a robust, easy-to-use, scalable simulation framework for large-scale multiphysics PDE-based network applications, integrating composable mathematics solvers and topology management. We have applied DMNetwork to the U.S. River network with millions of reaches and junctions on extreme scale supercomputers. Last, I will present our ongoing simulation of traffic flows using graph neural network (GNN) machine learning technique.